Image segmentation based on improved fuzzy clustering algorithm

To avoid the over and under segmentation problem in image segmentation, taking advantage of fuzzy clustering which is unsupervised and the simulated annealing principle can seek the optimal solution automatically, an approach for automatically image segmentation using improved fuzzy clustering algorithm based on the simulated annealing principle and the reversible jump Markov chain is proposed. First, the spatial information and the color information are considered to acquire the feature vectors of each pixel. Then by using the cluster validity index as the performance indicators and iteratively updating the segmentation number based on different moves, such as birth, death, split, merge, and perturb move. Finally, the simulated annealing principle was applied to seek the most suitable segmentation number, which can get more accurate and reasonable segmentation results automatically without prior knowledge or complex pretreatment. The experimental results show the proposed method can accomplish the image segmentation effectively and robustly.